Sierra Publishes Draft of Poppy Protocol, Adds 35 More Partners

On October 9, 2026, Sierra AI published a draft of the Personal Agent Protocol, which the company calls Poppy, and announced 35 additional design partners. That is almost all the confirmed information: the announcement consists of two facts, and both concern the project's status, not its technical design. So the most honest approach is to sort out what is known, what can reasonably be assumed, and what questions remain open.

What is confirmed and what is not

The announcement says little, but what it says is definite:

• Name: Personal Agent Protocol, short name Poppy.
• Status: a draft has been published, which means the specification may change substantially before the final version.
• Participants: 35 new design partners. The word "additional" means there were partners before, but the text gives neither their number nor a list.
• Author: Sierra AI, published on October 9, 2026.
• Not disclosed: the license, message format, authentication model, governing bodies for the standard, and the timeline for version 1.0.

Everything else in this piece is interpretation, and we label it as such.

Why Sierra wants a protocol

Sierra is known for building AI agents for businesses, primarily for customer service. It was founded by Bret Taylor and Clay Bavor. Today such an agent sits on the company's side: it answers the customer, processes a refund, changes a plan.

The name Personal Agent Protocol hints at the other side of the conversation. If a user approaches a business not directly but through their own agent, the two sides need a common language: how the agent introduces itself, what authority the person has granted it, which actions it may take without confirmation, and how the company verifies that it is not dealing with a fake. This is an inference from the name, not a quote from the specification. But the problem is real, and right now everyone solves it in their own way.

Other protocols already exist nearby. Anthropic's MCP describes how a model accesses tools and data. Google's A2A defines how agents interact with each other. There are also separate initiatives for payments in agentic commerce. Judging by its name, Poppy is going after the niche of "a person's agent versus a company's agent." Whether it will overlap with existing standards or interlock with them is not clear from the draft yet.

Why 35 partners matter

Protocols have no value on their own. Their value grows with the number of parties that implement both sides of the connection. At an early stage, design partners do three things at once: they bring real-world scenarios, find holes in the specification, and give the first implementations a chance to converge with one another.

The figure of 35 suggests that Sierra intends to build an ecosystem, not just publish a document. But it also raises the question of governance. If one company maintains the specification, partners gain influence over the details but not over the standard's fate. Whether Sierra will hand Poppy to a neutral foundation, open up the license, or let competitors into the discussion will become clear later. The answer determines whether we will see a second major player among the implementations.

What teams should do now

A practical course of action looks like this.

• Read the whole draft and note how it handles delegation of authority, confirmation of risky actions, auditing, and liability for an agent's errors. These four topics determine whether the protocol is suitable for money and personal data.
• Do not build production integrations on a draft. A prototype is acceptable; commitments to customers are not.
• Compare Poppy against your current stack: if you use MCP or A2A, check whether it complements them or requires a separate layer.
• Sierra customers should ask their account manager about support timelines and whether Poppy will be included in the product.
• Those building consumer agents would be wise to think ahead about how their agent will prove its authority to a company. This task will arise regardless of the fate of any particular protocol.

For now, this is a bid for a standard, not the standard itself. What will decide the outcome is the number of real implementations, not the number of partners in a press release.